Panyam Sri Naga Karthikeya Goud
Panyam Sri Naga Karthikeya Goud

Panyam Sri Naga Karthikeya Goud

AI/ML • Senior • Available in 2 weeks
Software Engineering Intern
2+ projects
Java
SQL
TypeScript
Node JS
Python
Data Pipeline & Feature Engineering: 4/10
MLOps & Deployment: 4/10
Active 4 days ago
Invite to interview
Message
Download CVCV
Overview
Technical skills
Timeline
Roles

Overview

RAG-focused LLM engineer at a senior level specializing in building modular retrieval-augmented generation systems and small ML prototypes. The strongest proven skill is designing a modular RAG pipeline and vector DB abstraction, evidenced by document_rag/rag.py and the vector_db implementations (vector_db/base.py and vector_db/qdrant.py). There is limited evidence of large-scale production deployment, GPU/efficiency engineering, or original research-level model development.

Technical skills

Java
SQL
TypeScript
Node JS
Python• Senior
Java
Spring Boot
Node JS
Axios
Python
Pydantic
Databases
Qdrant
MySQL• 2 projects
PostgreSQL
AI/ML
Deep Learning
TF-Keras
Transformers
RAG
OpenAI SDK
NumPy
Pandas
Frontend
React.js
Next.js
Tailwind CSS
Vite
Three.JS
GSAP
DevOps
AWS
Rest API

Timeline

Software Engineering Intern • Junior
Dataflow Group • Internship
In office
Performed security vulnerability remediation through VAPT, addressing issues such as stored XSS, injection vectors, missing rate limiting, and unsafe file upload handling. Implemented context-handling logic to reduce token usage while maintaining response quality. Shipped full-stack feature work for a live production application, including a workflow for verification requests with routing, SLAs, and automated email notifications, plus ongoing bug fixes and enhancements.
Senior AI/ML Engineer Confidence: Medium LLM Engineer
RAG-focused LLM engineer at a senior level specializing in building modular retrieval-augmented generation systems and small ML prototypes. The strongest proven skill is designing a modular RAG pipeline and vector DB abstraction, evidenced by document_rag/rag.py and the vector_db implementations (vector_db/base.py and vector_db/qdrant.py). There is limited evidence of large-scale production deployment, GPU/efficiency engineering, or original research-level model development.
Model Architecture & Training
2/10
How well models are designed and trained
Model architecture and training work is mostly application-level: standard LSTM for audio and TensorFlow Object Detection transfer-learning pipeline edits; no custom architectures, custom losses, or advanced training loops.
Evidence
Speech-Emotion-Recognition/speech emotion recognition.ipynb: Sequential LSTM model definition and model.fit training loop
Gesture-Glance/Tutorial.ipynb: Transfer-learning edits to TensorFlow Object Detection pipeline.config (pipeline_config.model.ssd.num_classes, fine_tune_checkpoint)
Data Pipeline & Feature Engineering
4/10
How data is prepared for models
Reasonable data pipeline and preprocessing engineering: TFRecord creation from Pascal VOC XML, MFCC extraction for audio, and PDF text extraction with chunking and overlap to prepare RAG inputs.
Evidence
Gesture-Glance/Tensorflow/scripts/generate_tfrecord.py: xml_to_csv, create_tf_example and TFRecord writer
Speech-Emotion-Recognition/speech emotion recognition.ipynb: extract_mfcc function and dataset loading/preprocessing
Document_RAG/document_rag/vector_db/base.py: read_pdf_document with chunk_size, chunk_overlap and text preprocessor/encoder hooks
Experimentation & Evaluation
3/10
How results are measured and tested
Basic experimentation and evaluation pipeline exists in notebooks and tests, including training/validation splits, plotting, confusion matrices, and small unit tests for components; lacks centralized experiment tracking or rigorous reproducibility setup.
Evidence
Speech-Emotion-Recognition/speech emotion recognition.ipynb: training history, accuracy/loss plots, confusion matrix and classification_report
Document_RAG/tests/test_rag.py and tests/test_llm.py: unit tests for RAG and LLM loading/predict behavior
MLOps & Deployment
4/10
How models are shipped to production
Deployment and MLOps concerns are addressed at an application level: a CLI ingestion/chat loop, pydantic settings for env-driven config, and a Qdrant vector DB adapter; however there are no production-grade serving patterns, scaling or monitoring integrations.
Evidence
Document_RAG/chatbot.py: CLI ingestion flow, vector DB cache clearing, interactive prompt loop
Document_RAG/document_rag/settings.py: Settings class (pydantic BaseSettings) with env-driven configuration
Document_RAG/document_rag/vector_db/qdrant.py: QdrantVectorDB adapter implementing add_documents and search
Computational Efficiency
1/10
How efficiently computing resources are used
Almost no evidence of focused computational efficiency engineering such as quantization, batching abstractions, GPU profiling, FlashAttention, or Triton kernels; only minimal batch size tuning in pipeline edits and standard model.fit usage.
Evidence
Gesture-Glance/Tutorial.ipynb: pipeline_config.train_config.batch_size change
Speech-Emotion-Recognition/speech emotion recognition.ipynb: standard model.fit with batch_size parameter
Research Depth & Innovation
1/10
Depth of research and new ideas
No original research contributions or novel model designs; the projects reuse established models and toolchains (transformers, CrossEncoder, SSD MobileNet, LSTM) without paper-reproduction artifacts or custom layers.
Evidence
Document_RAG/document_rag/llm/huggingface.py and ranker/huggingface.py: wrappers around existing HuggingFace components
Gesture-Glance/Tutorial.ipynb: uses SSD MobileNet from TensorFlow Model Zoo for transfer learning
Expertise
RAG• Senior
LLM• Middle
Technologies
Deep Learning
Python• Senior
Qdrant
OpenAI SDK
Transformers
Pandas
NumPy
RAG
TF-Keras
Pydantic
TensorFlow• mentioned only
Recommendations
  • Build production-ready serving and monitoring for the RAG system - add a REST/gRPC service, request/response latency SLOs, and basic drift logging.
  • Add reproducible experiment tracking (W&B/MLflow) and structured configs for training runs to improve comparability and auditability.
  • Implement batching/encoder optimization and memory profiling for embedding generation to reduce ingestion time and costs when scaling document ingestion.
  • Expand automated tests to include integration tests for vector DB + ranker + LLM end-to-end behavior and CI hooks for pre-commit checks.
Repositories
The developer's experience in this domain has been verified based on AI analysis of the following repositories:
Gesture-Glance Speech-Emotion-Recognition Document_RAG
Middle Backend Developer Confidence: Medium Data Platform
A middle-level backend developer focused on retrieval-augmented generation (RAG) and vector-search toolchains. The strongest proven skill is designing modular RAG pipelines and vector-DB integrations as implemented in document_rag/rag.py and the BaseVectorDB / QdrantVectorDB adapters. There is limited evidence of production-grade operations, distributed scaling, API versioning or advanced security/hardening in the public code.
API Design
2/10
How well APIs are designed
Minimal external API design; library-style, in-process APIs for LLM/RAG are present but no HTTP/versioning/idempotency or error-contract strategy for production APIs is implemented.
Evidence
Document_RAG/document_rag/rag.py: RAG.generate signature and PROMPT_TEMPLATE
Document_RAG/document_rag/llm/base.py: BaseLLM.generate interface
Document_RAG/document_rag/llm/openai.py: OpenAILLM.generate wrapper
Data Layer & Database
4/10
Working with databases
Deliberate data-layer abstractions and ingestion logic for vector search with clear chunking, metadata and a Qdrant adapter; migration history or DB transaction/isolation discussion is absent.
Evidence
Document_RAG/document_rag/vector_db/base.py: read_pdf_document, chunking, BaseVectorDB interface
Document_RAG/document_rag/vector_db/qdrant.py: QdrantVectorDB.create/add_documents/search implementations
Document_RAG/document_rag/settings.py: CHUNK_SIZE, CHUNK_OVERLAP and vector DB cache config
Scalability & Performance
3/10
Handling load and speed
Some performance-aware choices (chunking, ranker+retriever two-stage pipeline, numpy for scoring) but no measured optimizations, caching/invalidation, queueing or load-control in production.
Evidence
Document_RAG/document_rag/rag.py: uses retriever_chunks + ranker_chunks and numpy.argsort for ranking
Document_RAG/document_rag/vector_db/base.py: CHUNK_SIZE/CHUNK_OVERLAP constants and comment about parallelizing PDF extraction
System Architecture
4/10
Overall system structure
Clean modular architecture - clear interfaces and factories (LLM, ranker, vector DB) and a small orchestration class (RAG) that composes them; it is a single-process library without service decomposition or ops wiring.
Evidence
Document_RAG/document_rag/rag.py: class RAG and RAG.from_settings factory
Document_RAG/document_rag/llm/__init__.py: load_llm / BaseLLM abstraction
Document_RAG/document_rag/vector_db/base.py: BaseVectorDB abstract class and create_vector_db pattern
Security & Auth
2/10
Protecting data and access
Basic secrets/config handling using pydantic settings is present and there are simple input checks, but there is little evidence of runtime secrets hygiene, token lifecycle management, or explicit SQL/SSRF/SQLi hardening beyond basic validation.
Evidence
Document_RAG/document_rag/settings.py: Settings class exposing env-configured API keys and vector DB cache path
Document_RAG/document_rag/vector_db/base.py: read_pdf_document checks for file extension and existence
Document_RAG/document_rag/vector_db/qdrant.py: search raises ValueError when DB is empty
Reliability & Observability
2/10
Stability and monitoring
There is unit test coverage for core components and explicit error handling in the RAG flow, but minimal operational observability, retries/backoff, timeouts, structured logging or graceful shutdown patterns for production services.
Evidence
Document_RAG/tests/test_rag.py: tests for RAG behavior (add_pdf_documents, generate)
Document_RAG/tests/test_llm.py: tests for LLM loading and generate behavior
Document_RAG/document_rag/rag.py: exception paths and error-returning behavior when vector DB/ranker/LLM are used
Expertise
Backend AI & LLM• Middle
Databases & Vector Storage• Middle
Industries
Artificial Intelligence• Middle
Technologies
Java
SQL
Rest API
Spring Boot
AWS
Recommendations
  • Build RAG/LLM pipelines and vector-store integrations (ingest, chunking, ranker, LLM) using the existing document_rag abstractions.
  • Implement production hardening for vector search services: timeouts, retries with backoff, structured logging and metrics for the ingestion and query paths.
  • Extend the vector DB adapter layer with pluggable persistence strategies and a migration/versioning story for embeddings and metadata.
  • Develop test-driven data-ingestion pipelines for PDF and other document sources with parallelization and memory profiling for large-scale ingestion.
Repositories
The developer's experience in this domain has been verified based on AI analysis of the following repositories:
vocallabs Document_RAG Portfolio
Intern Frontend Developer Confidence: Low Generalist
Junior frontend developer focused on modern JavaScript frontends and build tooling with a preference for React/Next.js. The clearest, verifiable artifact is package.json evidence of Next.js, React and TypeScript usage which supports a basic Next/Vite frontend toolchain. There is little validated original, high-ownership application code exposed, and no strong public evidence of cross-cutting system design, operationalization or sustained production ownership.
UI Component Architecture
How interface parts are built
Not evidenced in public code
Responsive & Cross-browser
Works on all screens and browsers
Not evidenced in public code
Performance Optimization
Speed of the interface
Not evidenced in public code
Accessibility & Semantics
Usable for everyone
Not evidenced in public code
State Management & Data Flow
Managing data in the app
Not evidenced in public code
UX & Visual Polish
Look and feel quality
Not evidenced in public code
Expertise
React• Intern
Modern Web Frameworks• Intern
Frontend Architecture & Build Tools• Intern
Technologies
TypeScript
Node JS
GSAP
Tailwind CSS
Next.js
Three.JS
React.js
Vite
Axios
Recommendations
  • Develop and maintain user-facing React/Next.js features and small pages using the existing Next/Vite stacks.
  • Implement and polish interactive portfolio sections or marketing pages that use Three.js/GSAP where visual polish is required.
  • Ship UI component work and small integrations (Axios GraphQL/REST calls, auth hooks) under senior guidance to harden async and error handling.
  • Work on converting template code into owned components and add test coverage and CI gating to prove production ownership.
Repositories
The developer's experience in this domain has been verified based on AI analysis of the following repositories:
vocallabs Portfolio Swasthiq_Assignment